Beruflich Dokumente
Kultur Dokumente
Table 1
SDR to DDR Quick Reference
PARAMETER SDR DDR NOTES
DQM Yes No Used for write data mask and read OE
DM (Data Mask) No Yes Replaces DQM, used to mask write data only
DQS (Data Strobe) No Yes New, used to capture data
CK# (System Clock) No Yes New, DDR utilizes differential clocks
VREF No Yes Reference voltage for differential inputs (1/2 VDD)
VDD and VDDQ 3.3 Volts 2.5 Volts Reduced supply and power for DDR
Signal Interface LVTTL SSTL_2 DDR utilizes differential I/O
Output Drive Fixed Variable x16 DDR devices offer a reduced drive option
Data Rate 1x Clock 2x Clock Data transfer is twice the clock rate for DDR
Architecture Synchronous Source-Synchronous DDR utilizes a bidirectional data strobe
Figure 1
Functional Block Diagram
2 Meg x 4 Memory Array with SDR and DDR Interface
CKE
CK# 1 1
DQM
CK
SDR I/O Interface
CS# CONTROL
COMMAND
LOGIC
DECODE
WE# DATA
Generic Memory Array 4 OUTPUT
BANK3
CAS# BANK2 DQM REGISTER
RAS# BANK1 Not Used for DDR DQM
(SDR) DQ0-
Internal Data Bus 4 4 DQ3
REFRESH DATA
12
MODE REGISTERS COUNTER INPUT
ROW- 12 BANK0
ROW- REGISTER
ADDRESS BANK0
MUX ADDRESS MEMORY 4
12 4,096
LATCH ARRAY
12 AND (4,096 x 1,024 x 8)
DECODER
SENSE AMPLIFIERS
CK
4,096 DDR I/O Interface
DATA DLL
Internal Data Bus
4
2
I/O GATING x4 for SDR 8 4
READ
x8 for DDR MUX
A0-A11, ADDRESS BANK LATCH 4 DRVRS
BA0, BA1 14
REGISTER CONTROL DQS 1
2 LOGIC GENERATOR
1,024 DQ0-
(x8) DQ3, DM
Internal Data Bus COL0
DQS
(DDR) INPUT
8 REGISTERS
COLUMN DQS
DECODER 1 1
MASK
COLUMN- 1
ADDRESS WRITE 1 1
11 8 FIFO 2
COUNTER/
LATCH AND RCVRS
COL0 4 4
DRIVERS
8 4
1 Not Used for SDR 4 4
ck ck
out in DATA
COL0
CK
1
Figure 2
CAS#
Simplified Block Diagram
of 2n-Prefetch READ
WE#
n-bit
data DQS
n-bit
Data
Register
2n-bit n-bit data
From data D0 plus DQS All DQ
A0–Ai CA
DRAM MUX Q and DQS
Core D1 C Outputs
n-bit
Data CLKD
A0–Ai RA
Register
n-bit DQS
data
EN AP
A10
To the user, from a high-level view, 2n-prefetch DIS AP
means that data accesses occur in pairs; i.e., a single
read access fetches two data words; and for a single
BA0, 1 BA
write access, two data words (and/or data mask bits)
must be provided. This affects both the minimum burst
size and nonminimum burst interruptions. The mini- CA = Column Address
mum burst size of a 2n-prefetch architecture is two RA = Row Address
external data transfers. Ai = Most Significant Address
BA = Bank Address
EN AP = Enable Auto Precharge
Figure 3 DIS AP = Disable Auto Precharge
Simplified Block Diagram of
2n-Prefetch WRITE DON’T CARE
n-bit
data Minimum Time Slots
DQ0-DQi D Q
For READs, the controller can choose to ignore ei-
n-bit
Data ther of the two words, but the time slots for both will be
Register
occupied (see Figure 5). Similarly, for WRITEs, the con-
troller can mask either of the two words, but again, the
2n-bit time slots are occupied (see Figure 6). For each READ or
data
To
D Q D Q DRAM
WRITE command (and column address) applied, two
n-bit
Data
n-bit 2n-bit
Data
Core
data words are provided. Because the device is double
data
Register Register data rate as well as 2n-prefetch, a minimum of two data
DQS
words is optimal (since commands cannot be applied
CK more frequently).
Figure 5
Minimum Data Time Slot for 2n-Prefetch READ
CK#
CK
CL = 2
DQS
DO DO DO DO DO DO DO
DQ n n' x x' b b' g
For nonminimum READ bursts (four or eight words), Strobe-Based Data Bus
it helps to associate each positive clock edge with a pair In a purely synchronous system, data output and
of data words. This way, interruptions of READ com- capture are referenced to a common, free-running sys-
mands are easily understood. For example, with a burst tem clock. However, the maximum data rate for such a
length of eight, a READ command followed by three system is reached when the sum of output access time
uninterrupting commands is needed to access the en- and flight time approaches the bit time (the reciprocal
tire burst. If an interrupting command is applied at the of the data rate). Although generating delayed clocks
first positive clock edge following the READ command, for early data launch and/or late data capture will allow
only two words will be accessed; if an interrupting com- for increased data rate, these techniques do not ac-
mand is applied at the second positive clock edge fol- count for the fact that the data valid window (or data
lowing the READ command, only four words will be eye) moves relative to any fixed clock signal, due to
accessed, etc., (see Figure 7). changes in temperature, voltage, or loading. So, to al-
The concept of associating pairs of data with posi- low for even higher data rates, data strobe signals were
tive clock edges applies for WRITEs as well. However, to added to DDR devices. The data strobes are nonfree-
fully understand the masking and interrupting of write running signals driven by the device, which is driving
data, WRITE latency and strobe-based data bus tim- the data signals (the controller for WRITEs, the DRAMs
ing must be considered. For now it should be noted for READs). At the DRAM device level, for READs, the
that the positive clock edges of interest for WRITEs are data strobe (DQS) signals are effectively additional
different than those for READs. This is because of dif- data outputs (DQ) with a predetermined pattern; for
ferences in latencies and because the array access oc- WRITEs, the strobe signals are used as clocks to cap-
curs at the beginning of a READ operation but at the ture the corresponding input data. At the board level,
end of a WRITE operation. In fact, the relevant edges the strobe signals have identical loading to data sig-
depend on what the interrupting command is, as we nals and should be routed similarly.
will be shown later.
Figure 6
Minimum Data Time Slot for 2n-Prefetch WRITE
CK#
CK
tDQSS (NOM)
DQS
DI DI DI DI DI DI
DQ b x x' n' a a'
DM
At this point, it may be helpful to digress for a mo- For READs, the data strobe signals are edge-aligned
ment to cover the ratio of strobe-to-data signals. This with the data signals, meaning that all data and data
technical note focuses on typical desktop PC main strobes are clocked out of the device by the same inter-
memory applications, which use x64 DIMMs con- nal clock signal, and all will transition at the outputs at
structed with x8 or x16 DRAM devices. Controllers for nominally the same time. The controller will internally
these applications will be designed with one strobe per delay the received strobe to the center of the received
byte; i.e., there is one strobe on each x8 DRAM device data eye.
and two strobes on each x16 device. For WRITEs, the controller must provide the data
Other types of applications may use different strobe- strobes center-aligned relative to data. That is, strobe
to-data ratios. For example, servers using x72 DIMMs transitions occur nominally 90 degrees (relative to the
based on x4 components require controllers with one clock frequency) out of phase with data transitions.
strobe per four bits, and controllers for graphics or com- The DRAM device uses internally matched routing for
munications applications might use one to four strobes the strobes and data such that the strobes can be used
per 32 bits. Creative techniques can be utilized to mix directly to capture input data. Note that the reason
and match different strobe-to-data ratios in a system, READs and WRITEs use a different alignment scheme
but those will not be covered here. Because of the focus is so that the delay circuitry can be centralized in one
of this technical note, a strobe-to-data ratio of one per place (the controller) and does not have to be repli-
byte is used. The timing and calculations described cated in every DRAM device in the system. This ap-
apply independently to each group of signals. proach is expected to be carried forward to future gen-
erations of DDR to leverage the infrastructure now be-
ing established.
Figure 7
Relating Command Slots to Read Data
CK#
CK
CL = 2
DO
DQ n
CL = 2
DO
DQ n
CL = 2
DO
DQ n
CL = 2
DO
DQ n
Figure 8
DQS Pattern for READ Showing Preamble and Postamble
T0 T1 T2 T3 T4 T5 T6 T7 T8 T9 T10 T11
CK#
CK
Preamble Postamble
DQS
DO
DQ n
Figure 9
DQS Pattern for WRITE Showing Preamble and Postamble
T0 T1 T2 T3 T4 T5 T6
CK#
CK
Preamble Postamble
DQS
DQ DI
b
DON’T CARE
NOTE: 1. DI b = data-in for column b.
2. Three subsequent elements of data-in
are applied in the programmed order following DI b.
3. A burst of four is shown.
4. Shown with nominal tDQSS and without DM bits
for illustration.
Figure 10 Figure 11
Typical LVCMOS Receiver Typical SSTL_2 Receiver
VDDQ = 2.5V
VDDQ = 3.3V
VOUT
In Out VREF In
VSS
VSS
Figure 12
Typical SSTL_2 Interface and Input Levels
VIH(AC)
VTT
RT VIH(DC)
RS
VIN
+
VOUT VREF
-
VREF
Driver Receiver
VIL(DC)
VIL(AC)
Figure 13
Typical SSTL_2 Clocks
CK
1.45V X
1.25V VMP(DC) VID(DC)
VIX(DC) VID(AC)
1.05V X
CK#